Date post: | 01-Jan-2016 |
Category: |
Documents |
Upload: | carson-silva |
View: | 50 times |
Download: | 2 times |
http://www.youtube.com/watch?v=3EeJCln5KYg
+
Robotics
CS311, Spring 2013David Kauchak
Some material adapted from slides from Zach Dodds
+Admin
Assignment 5 graded
Exam #2 available later today To be done by Sunday at midnight
+What is a robot?
"I can't define a robot, but I know one when I see one.”--Joseph Engelberger (1966)
Justice Potter Stewart wrote in Jacobellis v. Ohio (1964), "I can't define pornography, but I know it when I see it."
Robot Defined
Word robot was coined by a Czech novelist Karel Capek in a 1920 play titled Rossum’s Universal Robots (RUR)
Robota in Czech is a word for worker or servant
Definition of robot:Any machine made by one our members: Robot Institute of America
A robot is a reprogrammable, multifunctional manipulator designed to move material, parts, tools or specialized devices through variable programmed motions for the performance of a variety of tasks: Robot Institute of America, 1979
Karel Capek
What is a Robot
Manipulator
What is a Robot
Wheeled RobotLegged Robot
What is a Robot
Unmanned Aerial VehicleAutonomous Underwater Vehicle
Robot Plot
Bar Monkey (9)
Roomba (7) Genghis (3)
Stanford Cart (3)Shakey (3)MERs (8)
Sims (5)
Capability (0-10)
Stanley/Boss (9)
Autonomyhuman-controlled independent
less
World Modeling
more
Unimate (4)da Vinci (2)
Robot timeline?
...1921 2421215020201950
Fictional Robot timeline
...1921 242121502020
Put these robots in chronological order?
Fictional robot timeline
...1921 2421215020201950
I, Robot
Karl CapekRossum’s Universal Robots
Asimov
Real robot timeline
...1951 1968 1976 1985
Real robot timeline
...1951
Tortoise “Elsie”
by Neurophysiologist Grey Walter
http://www.frc.ri.cmu.edu/~hpm/talks/revo.slides/1950.html
Shakey
Nils Nilsson @ Stanford Research Inst.
......1968
first “general-purpose” mobile platform
Living Room (L)
rem
sp
Kitchen (K)
Bedroom (B)
sh tv
Go(from,to)Preconditions: At(sh,from)
Postconditions: At(sh,to)
Push(obj,fr,to) Preconditions: At(sh,fr) At(obj,fr)
Postconditions: At(sh,to) At(obj,to)
Robotics's Shakey start
START
GOAL
ACTIONS
Go(L,B)
Go(L,K)
At(sh,L) At(sp,K) At(rem,B) At(tv,L)
Push(tv,L,B)
Push(tv,L,K)
At(sh,K) At(sp,K) At(rem,B) At(tv,K)
At(sh,L) At(sp,L) At(rem,L) At(tv,L)
+Shakey in video
http://www.youtube.com/watch?v=qXdn6ynwpiI
Stanford Cart: SPA
Hans Moravec @ SAIL
......1976
SE
NS
ING
AC
TIN
G
perc
eptio
n
wor
ld m
odel
ing
Pla
nnin
g
task
exe
cutio
n
mot
or c
ontr
ol
“functional” task decomposition“horizontal” subtasks
Cartland (outdoors)
Cartland (indoors)
“Robot Insects”
Rodney Brooks @ MIT
......1985
avoid objects
wander
explore
build maps
identify objects
planning and reasoning
SEN
SIN
G
AC
TIN
G
“behavioral” task decomposition“vertical” subtasks
+Robotics
What are the challenges?How do these relate to AI?
+AI
Search planning
Game playing
CSPs
Bayesian
HMMs
Machine learning neural nets
Knowledge representation
Natural Language processing
Computer vision
how much of the world do we need to represent internally ?
how should we internalize the world ?
what outputs can we effect ?
what inputs do we have ?
what algorithms connect the two ?
how do we use this “internal world” effectively ?
Autonomy/behavior
Robot Architecture
Robot Architecture
how much / how do we represent the world internally ?
Task-specific
Not at all
As much as possible!
Reactive paradigm
SPA paradigm
Behavior-based architecture
As much as possible.
Hybrid approaches
sense
plan act
history…
Sense - Plan - Act
......1976
SEN
SIN
G
AC
TIN
G
perc
epti
on
worl
d
modelin
g
pla
nnin
g
task
exe
cuti
on
moto
r co
ntr
ol
sense
plan act
Stanford Cart
Shakey
1968
MERs
… - 2009
Mars Exploration Rovers
Sense – Plan – Act "deliberative" architecture
Mars Science Lab
2011 - lasers, lifebio, and maybe nuclear-powered
Robot Architecture
how much / how do we represent the world internally ?
Task-specific
Not at all
As much as possible!
Reactive paradigm
SPA paradigm
Behavior-based architecture
As much as possible.
Hybrid approaches
sense
plan act
sense
act
stimulus - response
Biological Inspiration
Ethology: describing animal behavior
Getting to the ocean?
AI reasoning systems abstract too much away: frame problem
sense
act
Decision-making is based only on current sensor inputs.
“The world is its own best model”
Digger wasps’ nest-building sequence
Analog reactive robots
...1951
“Tortoise” Gray Walter
1984
“BEAM”Mark Tilden
“light-headed” behavior
stateless...
http://people.cs.uchicago.edu/~wiseman/vehicles/
1989-
Valentino Braitenberg
robot made from Playstation pieces…!
http://haroldsbeambugs.solarbotics.net/mercury.htm
commercial products…
Robot Architecture
how much / how do we represent the world internally ?
Task-specific
Not at all
As much as possible!
Reactive paradigm
SPA paradigm
Subsumption paradigmPotential Fields
Behavior-based architecture
As much as possible.
Hybrid approaches
sense
plan act
sense
act
different ways of composing behaviors
stimulus – response == "behavior"
......1985
avoid objects
wander
explore
build maps
identify objects
planning and reasoning
SEN
SIN
G
AC
TIN
G
Genghis
“Vertical” task decomposition
sense
act
little explicit deliberation except through system
state
Behavior-based control
Behavior a direct mapping of sensory inputs to a pattern of task-specific motor actions
extinguish approach wander
Subsumption
Subsumption builds intelligence incrementally in layers
runaway behavior
wander behavior
Subsumption
Where would a light-seeking behavior/layer connect?
runaway behavior
wander behavior
Subsumption
Where would a light-seeking behavior/layer connect?
runaway behavior
wander behavior
S
Closest LightLIGHT
SONAR
phototaxis
Subsumption - Limits
Success of behavior-based systems depends on how well-tuned they are to their environment. This is a huge strength, but it's also a weakness …
Herbert, a soda-can-collecting robot
Reaching the end of the subsumption architecture and
purely reactive approaches.
http://www.youtube.com/watch?v=YtNKuwiVYm0
Subsumption limits: Genghis
runaway behavior
wander behavior
navigate behavior
FSM / DFA
Unwieldy!
Larger example -- Genghis
1) Standing by tuning the parameters of two behaviors: the leg “swing” and the leg “lift”
2) Simple walking: one leg at a time
3) Force Balancing: via incorporated force sensors on the legs
4) Obstacle traversal: the legs should lift much higher if need be
5) Anticipation: uses touch sensors (whiskers) to detect obstacles
6) Pitch stabilization: uses an inclinometer to stabilize fore/aft pitch
7) Prowling: uses infrared sensors to start walking when a human approaches
8) Steering: uses the difference in two IR/range sensors to follow
57 modules wired together !
Robot Architecture
how much / how do we represent the world internally ?
Task-specific
Not at all
As much as possible!
Reactive paradigm
SPA paradigm
Subsumption paradigmPotential Fields
Behavior-based architecture
As much as possible.
Hybrid approaches
sense
plan act
sense
act
different ways of composing behaviors
Potential Fields
Potential fields compose simple behaviors by adding the outputs that each sensor/input sends the robot
Ron Arkin @ Georgia Tech
A sequencing process (FSM/DFA) updates the potential fields and/or decides which ones to run next…
Individual potential fields (motor schemas) contain state
Motor Schemas / Potential Fields
goal-seeking schemaobstacle-avoiding schema
note that the complete environmental vector fields are only for visualization!
Direct mapping from the environment to a control signal
combine?
Behavior Summer
vector sum of the avoid and goal motor schemas
path taken by a robot controlled by the resulting
field
Implementation details
the extent to which potential field force drops off with distance…
corridor-following schema(s)?what crucial assumption is being made here?
Additional behavior primitives
go! schemacorridor-centering schema
A more complex task
Direct mapping from the environment to a control signal
larger composite task
How many individual fields are summed in this task?
Not necessarily all at one time!
Local minima
A potential-field-based system can get stuck!
the problem
a solution?
What would happen if a robot came in in the middle on the left?
Local minima
A potential-field-based system can get stuck!
the problem
Why is the “local minimum” problem, as illustrated to the left, not likely to actually cause a robot to get stuck in practice?
robots controlled by summing goal/obstacle potential fields can get stuck in practice -- draw an example of an environment with both obstacle(s) and goals(s) in which getting stuck might actually occur.
Suggest how a robot might overcome the problem of getting stuck in such cases…
Local minima
A potential-field-based system can get stuck!
the problem a solution
Bigger deadends...
How to get out of larger wells ?
Bigger deadends...
uses memory of where the robot has been
past-avoiding motor schema
Another example
Keeping away from past locations...
Pfields in Practice
Steathy navigation @ USC (Ashley Tews, Gaurav S. Sukhatme, and Maja J. Mataric)
part of the potential field… What's going on here?
http://robotics.usc.edu/interaction/?l=Research:Projects:stealth:index#experiments
Docking with potential fields
How does the idea of docking, e.g., with an electrical outlet change the requirements for a potential field?
example goals
Why might a simple attractive force not be sufficient for docking (plugging-in, etc.)?
Docking with potential fields
The key insight is the need to establish an approach direction
example goals
Docking with potential fields
The key insight is the need to establish an approach direction
+Review
Machine learning general learning concepts
supervised vs. unsupervised features/feature-based problems/feature space bias/variance overfitting hyperplanes/linear seperability
Supervised learning applications approaches
k-NN decision trees NB SVM (large margin classifiers)
Ensemble approaches (boosting)
+Review
Machine learning (continued) unsupervised learning
application issues
number of clusters flat vs. hierarchical soft vs. hard clustering
approaches k-means EM
word alignment clustering (mixture of gaussians)
spectral clustering (min-cut)
+Review
Neural networks (Machine learning?) perceptrons/neurons
activation functions (threshold vs. sigmoid) perceptron learning
multi-layer networks
Knowledge representation basic logic ontology NELL
+Review
CSPs problem formulation
variables domain constraints
why CSPs? applications? constraint graph CSP as search
backtracking algorithm forward checking arc consistency
heuristics most constrained variable least constrained value ...
+Review
Natural language processing Applications Problem areas Why it’s hard? Machine translation setup
+Guest speaker
Rodney Brooks Professor at MIT (was previous director of CSAIL) Founder of iRobot
http://www.youtube.com/watch?v=B79D9nW2AFA